Prometheus PSR Details

Retail Store Agentic Topos World Model

Readable local predictive-state details for the product-feedback Topos World Model. This view exposes the finite histories, tests, restriction checks, gluing overlaps, and j-do probes that sit underneath the persistent state.

Raw world-model JSON

Model State

review episodes400032,000 events

Predictive-state contexts17finite context family

Restriction arrows26/3416 root + 18 local

Mean glue loss0.0975GB diagnostic

Overlap edges6learned overlap graph

j-do probes0local intervention tests

Predictive-State Contexts

corpus

All extracted causal observation events.

Events Rank 22Histories 1004Tests 2003

agent:buyer

Local context with 1000 extracted events.

Events 1000Rank 1Histories 1Tests 979

agent:clerk

Local context with 1000 extracted events.

Events 1000Rank 1Histories 1Tests 979

agent:scheduler

Local context with 1000 extracted events.

Events 1000Rank 1Histories 1Tests 979

agent:supervisor

Local context with 1000 extracted events.

Events 1000Rank 1Histories 1Tests 979

guardrail:inventory mix

guardrail:inventory_mix

Local context with 4000 extracted events.

Events 4000Rank 1Histories 1Tests 2

guardrail:staffing

Local context with 4000 extracted events.

Events 4000Rank 1Histories 1Tests 4

payoff:checkout

Local context with 1000 extracted events.

Events 1000Rank 1Histories 1Tests 2

payoff:inventory

Local context with 1000 extracted events.

Events 1000Rank 1Histories 1Tests 2

payoff:staffing

Local context with 1000 extracted events.

Events 1000Rank 1Histories 1Tests 2

payoff:supervisor

Local context with 1000 extracted events.

Events 1000Rank 1Histories 1Tests 2

policy:online learning

policy:online_learning

Local context with 4000 extracted events.

Events 4000Rank 1Histories 1Tests 1000

role:checkout

Local context with 2000 extracted events.

Events 2000Rank 3Histories 5Tests 5

role:inventory

Local context with 2000 extracted events.

Events 2000Rank 5Histories 5Tests 7

role:staffing

Local context with 2000 extracted events.

Events 2000Rank 5Histories 5Tests 7

role:supervisor

Local context with 2000 extracted events.

Events 2000Rank 4Histories 5Tests 6

store:clearing

Local context with 4000 extracted events.

Events 4000Rank 1Histories 1Tests 3

Restriction Diagnostics

How to read this: a restriction compares two local PSR tables on shared history/test cells. Mean gap is the average absolute probability difference on those shared cells; max gap is the largest single-cell difference. Aligned means the local predictive behavior agrees with the source within tolerance; divergent marks a locally distinctive chart, not an error.

Restriction	Status	Shared cells	Mean gap	Max gap
corpus -> agent:buyer	aligned	979	0.0009	0.0016
corpus -> agent:clerk	aligned	979	0.0009	0.0016
corpus -> agent:scheduler	aligned	979	0.0009	0.0016
corpus -> agent:supervisor	aligned	979	0.0009	0.0016
corpus -> guardrail:inventory mix	divergent	2	0.4999	0.8488
corpus -> guardrail:staffing	divergent	4	0.2499	0.7826
corpus -> payoff:checkout	divergent	2	0.4999	0.6378
corpus -> payoff:inventory	divergent	2	0.4999	0.6378
corpus -> payoff:staffing	divergent	2	0.4999	0.6378
corpus -> payoff:supervisor	divergent	2	0.4999	0.6378
corpus -> policy:online learning	aligned	1000	0.0009	0.0009
corpus -> role:checkout	divergent	25	0.1677	0.9971
corpus -> role:inventory	aligned	35	0.1197	0.9957
corpus -> role:staffing	aligned	35	0.1197	0.9957
corpus -> role:supervisor	aligned	30	0.1397	0.9964
corpus -> store:clearing	divergent	3	0.3332	0.8667
agent:clerk -> agent:buyer	aligned	979	0	0
agent:scheduler -> agent:buyer	aligned	979	0	0
agent:supervisor -> agent:buyer	aligned	979	0	0
agent:scheduler -> agent:clerk	aligned	979	0	0
agent:supervisor -> agent:clerk	aligned	979	0	0
agent:supervisor -> agent:scheduler	aligned	979	0	0
payoff:inventory -> payoff:checkout	aligned	2	0	0
payoff:staffing -> payoff:checkout	aligned	2	0	0
payoff:supervisor -> payoff:checkout	aligned	2	0	0
payoff:staffing -> payoff:inventory	aligned	2	0	0
payoff:supervisor -> payoff:inventory	aligned	2	0	0
payoff:supervisor -> payoff:staffing	aligned	2	0	0
role:inventory -> role:checkout	aligned	20	0.0064	0.0317
role:staffing -> role:checkout	aligned	20	0.0064	0.0317
role:supervisor -> role:checkout	aligned	20	0.0036	0.0179
role:staffing -> role:inventory	aligned	20	0	0
role:inventory -> role:supervisor	aligned	20	0.0028	0.0139
role:staffing -> role:supervisor	aligned	20	0.0028	0.0139

Gluing Diagnostics

How to read this: gluing compares projected latent sections for overlapping contexts. Weighted loss combines geometric mismatch with overlap confidence, so high-loss rows point to places where local charts may need separate narrative treatment.

Overlap	Status	Sections	Weighted loss	Max gap
role:checkout -> role:supervisor	divergent	5	0.1978	0.4349
role:inventory -> role:supervisor	divergent	5	0.1782	0.3721
role:checkout -> role:staffing	divergent	5	0.16	0.3103
role:staffing -> role:supervisor	divergent	5	0.1318	0.2482
corpus -> guardrail:inventory mix	divergent	1	0.1231	0.1403
role:checkout -> role:inventory	divergent	5	0.1135	0.2399
corpus -> payoff:checkout	divergent	1	0.0833	0.0909
corpus -> payoff:inventory	divergent	1	0.0833	0.0909
corpus -> payoff:staffing	divergent	1	0.0833	0.0909
corpus -> payoff:supervisor	divergent	1	0.0833	0.0909
role:inventory -> role:staffing	divergent	5	0.0817	0.1713
corpus -> guardrail:staffing	aligned	1	0.0797	0.0866
corpus -> role:checkout	aligned	5	0.0729	0.1256
corpus -> role:supervisor	aligned	5	0.072	0.1231
corpus -> role:inventory	aligned	5	0.0714	0.1221
corpus -> role:staffing	aligned	5	0.0711	0.1208

Matrix Preview

store:clearing	residual:low\|lost:low\|stockout:candle	residual:low\|lost:low\|stockout:none	residual:medium\|lost:low\|stockout:none
<START>	0.1321	0.8668	0.0011

Showing up to 8 histories and 6 tests. The complete finite predictive-state family is preserved in the world-model JSON.

j-do Probes

No j-do probes are available.